[yusufg at outblaze.com: ext3 reservation seems to cause major slowdown in synctest in 2.6.10-rc2 vs 2.6.9]

[Yusuf Goolamabbas]

Maybe of relevance to fedora developers since ext3 reservation patch was
first introduced in Fedora. Not sure how close the current patch in FC3
is to the one in 2.6.10-rc2

Would appreciate if others could try the same benchmark with different
i/o controllers (ATA/SATA/3ware/Fibre)

----- Forwarded message from Yusuf Goolamabbas <yusufg at outblaze.com> -----

Date: Mon, 15 Nov 2004 14:33:25 +0800
From: Yusuf Goolamabbas <yusufg at outblaze.com>
To: linux-kernel at vger.kernel.org
Subject: ext3 reservation seems to cause major slowdown in synctest in 2.6.10-rc2 vs 2.6.9 
Message-ID: <20041115063325.GA31537 at outblaze.com>

I have a P3-500 box /384 MB ram and 2 scsi disks (sda and sdb).
OS is on sda and test partions is on sdb
aic7xxx driver 
Using anticipatory schedulor with a tag_depth of 4 (this is set via
modules.conf)
options aic7xxx aic7xxx=global_tag_depth:4

/dev/sdb1 is created with ext3 and htree is enabled. Mounted as /htree

synctest obtained from here (synctest tries to simulate an MTA
behaviour)

http://www.zip.com.au/~akpm/linux/patches/stuff/ext3-tools.tar.gz

I run the following commands on both 2.6.9 and 2.6.10-rc2

/usr/bin/time -p ./synctest -fu -t 100 -p1 -n1 /htree/nfsexport

timing results

2.6.9 

real 57.35
user 1.37
sys 14.26

2.6.10-rc2

real 86.83
user 1.32
sys 14.02

Mounting with noreservation gives the following numbers

real 58.77
user 1.46
sys 14.48


----- End forwarded message -----